Is there a reason OpenAI and Anthropic are focusing on solving abstract mathematical problems, where as Tao says, the results themselves are not the goal, rather than curing cancer, where results matter more than understanding?

Sep 12, 2026 · 5:49 AM UTC

59
7
288
28,982
Lots of interesting responses here, one point I saw elsewhere on my feed (might have missed it here): showing superiority in math is relevant to the goal of being adopted and funded by the military-industrial complex
1
1
11
814
The cynic in me also wonders: one they can sell this as "AGI", would the incentives to perform rigorous biological testing remain? "AI predicted this would cure disease X" might be enough advertising for people to start taking the drug.
1
5
730
Sort replies: Relevant Recent Liked
Replying to @postdocforever
cause the models are not good at that yet (hopefully they will be soon). the only area they are superhuman currently in, and therefore can produce groundbreaking results, is this.
3
13
2,056
It's also a question of aiming for it no? They have been employing mathematicians to improve models and strategies to pursue this, so of course they are making progress
1
5
1,843
more replies
Replying to @postdocforever
I believe that addressing those "big problems", such as curing cancer, will likely only be possible once we have developed a solid understanding of crucial maths issues. modeling anomalous cancer growth will depend on our understanding of unresolved questions regarding maths
3
5
2,162
I don't fully agree. That would depend on us having an actually good mathematical model of the biology, which we do not. We have various approximate theories, and also the math for solving them is distant from methods used to prove theorems in pure math afaik.
1
34
1,845
more replies
Replying to @postdocforever
Incentives
64
Replying to @postdocforever
Yes. Finding a cure for cancer is harder, and it’s unlikely LLMs will help much.
201
Replying to @postdocforever
in the actual sciences, they are experiment-bound. only so much you can do with protein folding calculations.
10
283
Replying to @postdocforever
They arent solving abstract problems. They are cherry picking the problems that can be solved with brute force computation without any abstract understanding
1
1
5
180
Replying to @postdocforever
Marketing. Mathematics is culturally associated with two things: 1) intellectual rigour 2) absolute truth. If an AI is seen to have access to these things, then it is seen as more powerful and hence becomes more saleable.
4
156
Replying to @postdocforever
Don't be naive, OpenAI will 100% steal a cure for cancer when the opportunity arises.
5
105
Replying to @postdocforever
Perhaps because due to proof-checking languages they can know almost immediately whether their proof is correcte. But should they invent a new drug against cancer, it could take years to ensure it is efficient and does not cause too much collateral damages.
4
191
Replying to @postdocforever
because 1. trying to "cure cancer" is almost like asking to "cure virus" 2. they have a machine that can grind through solutions to problems that can exist in the digital space, biological problems exist in the real world and need experiments to solve them.
1
1
101
Replying to @postdocforever
very difficult to cure cancer. Very feasible to solve Clay Prize. Low hanging fruit.
2
60
Replying to @postdocforever
Lean. Math problems are purely intellectual, so no labs or empirical work required, and with Lean we have an objective way of telling whether a proof is correct. It's the same thing with coding, you only need a computer and check whether it compiles/passes tests.
2
287
Replying to @postdocforever
Because they are not solving anything, they are selling. They only need the easy news/post and the noise on social networks
2
26
Replying to @postdocforever
math is a good place for ai to self improve very efficiently, probably one of the few ones tbh
1
156
Replying to @postdocforever
Because tao is wrong and the results themselves ARE the goal.
1
1
252
Replying to @postdocforever
But Tao isn’t completely right. Meaningful theorems are major goals in mathematics. They do matter.
1
30
Replying to @postdocforever
It's where they able to make the biggest advances now. Math follows explicit rules and the AI algorithms and massive computing power they have are well-suited to proving theorems.
1
359
Replying to @postdocforever
Mathematics don't need physical world, math books and papers contain millons of problems and solution. Easy to access all mankind math knowledge and now AI labs tried to solver the most famous and hardest math problems as benchmark. AI for cancer need real labs with real cells.
1
14
Replying to @postdocforever
Cancer research is very regulated, math more free
1
1
150
Replying to @postdocforever
Marketing. Same reason they probably locked a few hundred or thousand people into a big room and had them post-train Astra on how to use Blender better.
1
150
Replying to @postdocforever
Cause math/theoretical physics is ultimate intellectual exercise. Mathematicians/Physicists are smartest people on earth If they can pretend ai can replace mathematicians/physicists by definition they can pretend ai can replace software engineers etc.(which are 2 3 levels below)
1
254
Replying to @postdocforever
Because the only way they know how to check correctness is with an external fitness test. Only pure math has that, not even software does. They literally can't focus on anything else. It's a party trick
1
142
Replying to @postdocforever
Symbolic nature of problem makes it easier & hence quicker to verify. Pretty simple Curing cancer altho would be easily EV+ on any reasonable horizon, would still take years to verify In mean time if replacing math PhDs is on same critical line & helps raise capital, why not?
207
Replying to @postdocforever
The ai can come up with candidate drugs but that isn’t the bottleneck in cancer research
1
25
Load more